<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova" -->

---
title: Confidence is not transferable | daily.dev
description: Building on a prior experiment where an AI agent implemented a complex code metric in 15 minutes but left the author unable to verify correctness for hours,...
canonical: https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Confidence is not transferable | daily.dev
og:description: Building on a prior experiment where an AI agent implemented a complex code metric in 15 minutes but left the author unable to verify correctness for hours,...
og:url: https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova
og:image: https://api.daily.dev/og/posts/LsMgmjOVa.png
og:image:alt: Confidence is not transferable
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Confidence is not transferable

**[PHPUnit Expert](https://daily.dev/sources/phpunit-expert)** · 8 min read · 8 upvotes · 2 comments

## Summary

Building on a prior experiment where an AI agent implemented a complex code metric in 15 minutes but left the author unable to verify correctness for hours, this piece draws on three external observations (from Alberto Brandolini, Zsófia Herendi, and Andreas Bulling) covering organisational, individual, and generational angles on AI-assisted development. It introduces the idea of 'borrowed confidence' (or 'confidence laundering'): AI produces plausible-looking code, tests, and explanations, but none of these signals constitute independent evidence of correctness. The piece argues this asymmetry hits differently by career stage, with newcomers who never earned confidence at greatest risk. It then describes three concrete techniques for earning independent confidence in AI-generated code: a brute-force oracle for differential testing, metamorphic relations derived from a specification, and mutation testing of AI-generated test suites. It concludes that the scarce future skill is producing justified confidence in code, reframing test-driven development as deliberate practice for judgement rather than just bug-catching.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://phpunit.expert/articles/confidence-is-not-transferable.html>

## Questions this post answers

### How can I verify AI-generated code is correct when I don't have deep expertise in the algorithm's domain?

Independent evidence is needed rather than trusting the AI's own tests, since passing tests only show the implementation agrees with itself. Three techniques help: a brute-force oracle (a simple, obviously-correct implementation checked against the AI's version via differential testing), metamorphic relations derived from the specification rather than the code, and mutation testing to check whether the AI-generated test suite actually detects injected defects.

_Developers verifying AI-generated code can find testing techniques like these curated on daily.dev._

### What is confidence laundering in the context of AI-generated code?

Confidence laundering describes how plausibility goes in and apparent certainty comes out of an AI coding process, without independent evidence ever entering the loop. Clean-looking code, passing tests, plausible visualizations, and articulate explanations are all generated by the same agent that wrote the implementation, so they only prove self-consistency, not correctness against an external specification.

_Teams weighing how much to trust AI-written code and tests follow discussions like this on daily.dev._

### Why does test-driven development still matter when AI can generate code and tests quickly?

Test-driven development was never primarily about catching bugs; writing the test first forces defining expected behavior before a solution exists to seduce you into skipping that judgment. It functions as deliberate practice for judgment, the kind of slow, friction-filled moment that AI-assisted workflows tend to optimize away, and that must now be preserved deliberately to keep earning real confidence rather than borrowing it.

_Practitioners rethinking TDD's role alongside AI coding agents track perspectives like this on daily.dev._

## Community discussion

Top comments from developers on daily.dev.

**@ankitqumar** · 0 upvotes

> IF TIME is Limited

**@nark3d** · 0 upvotes

> A brute-force oracle costs you every time the implementation moves.  Differential testing gets you a real signal, but if the oracle is a second implementation of the same spec then you've just moved the confidence problem sideways.  What makes it independent enough that a shared misunderstanding doesn't pass both?

## Similar posts on daily.dev

- [WHY WE TRUST AI-GENERATED CODE BEFORE WE VERIFY IT](https://daily.dev/posts/why-we-trust-ai-generated-code-before-we-verify-it-bmsbpoml4) · Medium · 0 upvotes · 0 comments
- [You Cannot Trust the AI Code Generator](https://daily.dev/posts/you-cannot-trust-the-ai-code-generator-schdivrx2) · RUBYLAND · 0 upvotes · 1 comments
- [Confidence Techniques for Guiding AI Migration](https://daily.dev/posts/confidence-techniques-for-guiding-ai-migration-erdpv4kbm) · Scott Logic · 0 upvotes · 0 comments
- [Why Confidence Without Visibility is a False Sense of AI Security](https://daily.dev/posts/why-confidence-without-visibility-is-a-false-sense-of-ai-security-dxdtj2bsu) · Security Boulevard · 0 upvotes · 0 comments
- [Agentic Skill Decay](https://daily.dev/posts/agentic-skill-decay-nzqkcham8) · Addy Osmani · 89 upvotes · 13 comments

---

Tags: [#testing](https://daily.dev/tags/testing), [#ai-coding](https://daily.dev/tags/ai-coding), [#tdd](https://daily.dev/tags/tdd), [#mutation-testing](https://daily.dev/tags/mutation-testing)

[View this post on daily.dev](https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Confidence is not transferable","url":"https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova"},"datePublished":"2026-09-16T04:35:29.643Z","dateModified":"2026-09-16T04:35:55.404Z","description":"Building on a prior experiment where an AI agent implemented a complex code metric in 15 minutes but left the author unable to verify correctness for hours,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/30f88886534b3dc2c77b34d34ab12f3e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/30f88886534b3dc2c77b34d34ab12f3e?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"PHPUnit Expert","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"PHPUnit Expert","logo":"https://media.daily.dev/image/upload/s--FMZoysoG--/f_auto,q_auto/v1774964057/logos/phpunit-expert?_a=BAMAMiWQ0","url":"https://daily.dev/sources/phpunit-expert"},"commentCount":2,"discussionUrl":"https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":8},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":2}],"keywords":"testing,ai-coding,tdd,mutation-testing","timeRequired":"PT8M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"PHPUnit Expert","item":"https://daily.dev/sources/phpunit-expert"},{"@type":"ListItem","position":3,"name":"Confidence is not transferable"}]}
{"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova","comment":[{"@type":"Comment","text":"IF TIME is Limited","datePublished":"2026-09-19T10:38:14.848Z","url":"https://daily.dev/posts/LsMgmjOVa#c-GEUyjktx6","author":{"@type":"Person","name":"ankit","url":"https://daily.dev/ankitqumar","image":"https://lh3.googleusercontent.com/a/ACg8ocKqYBZQkLczsLtAjJY-Xp0fEw7sAGL4WdVp5T6h2kcDwXqBIrw=s96-c"}},{"@type":"Comment","text":"A brute-force oracle costs you every time the implementation moves.  Differential testing gets you a real signal, but if the oracle is a second implementation of the same spec then you’ve just moved the confidence problem sideways.  What makes it independent enough that a shared misunderstanding doesn’t pass both?","datePublished":"2026-09-21T10:25:05.505Z","url":"https://daily.dev/posts/LsMgmjOVa#c-SRkxBemwF","author":{"@type":"Person","name":"Adam","url":"https://daily.dev/nark3d","image":"https://avatars.githubusercontent.com/u/2162621?v=4"}}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/confidence-is-not-transferable-lsmgmjova#faq","mainEntity":[{"@type":"Question","name":"How can I verify AI-generated code is correct when I don't have deep expertise in the algorithm's domain?","acceptedAnswer":{"@type":"Answer","text":"Independent evidence is needed rather than trusting the AI's own tests, since passing tests only show the implementation agrees with itself. Three techniques help: a brute-force oracle (a simple, obviously-correct implementation checked against the AI's version via differential testing), metamorphic relations derived from the specification rather than the code, and mutation testing to check whether the AI-generated test suite actually detects injected defects. Developers verifying AI-generated code can find testing techniques like these curated on daily.dev."}},{"@type":"Question","name":"What is confidence laundering in the context of AI-generated code?","acceptedAnswer":{"@type":"Answer","text":"Confidence laundering describes how plausibility goes in and apparent certainty comes out of an AI coding process, without independent evidence ever entering the loop. Clean-looking code, passing tests, plausible visualizations, and articulate explanations are all generated by the same agent that wrote the implementation, so they only prove self-consistency, not correctness against an external specification. Teams weighing how much to trust AI-written code and tests follow discussions like this on daily.dev."}},{"@type":"Question","name":"Why does test-driven development still matter when AI can generate code and tests quickly?","acceptedAnswer":{"@type":"Answer","text":"Test-driven development was never primarily about catching bugs; writing the test first forces defining expected behavior before a solution exists to seduce you into skipping that judgment. It functions as deliberate practice for judgment, the kind of slow, friction-filled moment that AI-assisted workflows tend to optimize away, and that must now be preserved deliberately to keep earning real confidence rather than borrowing it. Practitioners rethinking TDD's role alongside AI coding agents track perspectives like this on daily.dev."}}]}
```

